On Thu, 27 Apr 2017 22:49:51 -0700
Alexei Starovoitov <alexei.starovoi...@gmail.com> wrote:
> On Thu, Apr 27, 2017 at 01:15:42PM +0200, Jesper Dangaard Brouer wrote:
> >
> > To provoke this bug, remember that you MUST call:
> >
> > make headers_install
> >
> > In the kernels root directory, else you will be compiling samples/bpf/
> > against the older headers previously installed.
> >
> > The error looks like:
> >
> > $ sudo ./sockex1
> > bpf_load_program() err=22
> > fd 0 is not pointing to valid bpf_map
> > sockex1: [...]/samples/bpf/sockex1_user.c:26: main: Assertion
> > `setsockopt(sock, SOL_SOCKET, SO_ATTACH_BPF, prog_fd, sizeof(prog_fd[0]))
> > == 0' failed.
> > Aborted
> >
> > I've found that the bug were introduced in
> > commit: fb30d4b71214 ("bpf: Add tests for map-in-map")
>
> Great debugging!
> Indeed that change made samples/bpf/bpf_load.c to be incompatible with .o
> generated earlier. We should really get rid of that loader and
> switch to tools/lib/bpf/. I believe Eric Leblond already made it
> resilient to 'struct bpf_map_def' changes.
Yes, exactly it is problem in samples/bpf/bpf_load.c. As it assumes
the contents of the ELF file maps section will always chunks in
sizeof(struct bpf_map_def) and just uses that directly as a pointer to
an array of type struct bpf_map_def, which of-cause silently blows up
when changing struct bpf_map_def. That cost me many hours to discover
that yesterday.
I started implementing more correct parsing of the ELF maps section, it
is doable, but as you say, maybe we should just get rid of this loader?
I will at least fixup bpf_load.c and perhaps just abort the program the
program if I detect a difference between the ELF size and struct size.
And send this as a patch later today...
I've also looked at the loaded Daniel implemented[1] in iproute2, and
it is much cleaner.
[1]
https://git.kernel.org/pub/scm/linux/kernel/git/shemminger/iproute2.git/tree/lib/bpf.c
--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
LinkedIn: http://www.linkedin.com/in/brouer
